# US 2019/0377610 A1 · Determining an allocation of stage and destage tasks by training a machine learning module

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0377610 A1
- Document ID: US20190377610A1
- Status: Published application
- Kind: A1
- Application number: 16/137,233
- Filed: 2018-09-20
- Published: 2019-12-12
- Earliest claimed: 2018-06-08
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Lokesh M. Gupta; Kyler A. Anderson; Matthew G. Borlick; Kevin J. Ash
- Web page: https://patentgrep.com/patent/US20190377610A1

## CPC

- G06F 9/5038: {considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration (scheduling strategies G06F9/4881 and subgroups)}
- G06F 12/0868: Data transfer between cache memory and other subsystems, e.g. storage devices or host systems
- G06F 3/061: {Improving I/O performance}
- G06F 3/0659: {Command handling arrangements, e.g. command buffers, queues, command scheduling}
- G06F 3/0689: {Disk arrays, e.g. RAID, JBOD}
- G06F 9/4806: {Task transfer initiation or dispatching}
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 20/00: Machine learning
- G06F 2212/1024: Latency reduction
- G06F 2212/1041: Resource optimization
- G06F 2212/262: configured as RAID
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 5/025: Extracting rules from data
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Provided are a computer program product, system, and method for using a machine learning module to determine an allocation of stage and destage tasks. Storage performance information related to processing of Input/Output (I/O) requests with respect to the storage unit is provided to a machine learning module. The machine learning module receives a computed number of stage tasks and a computed number of destage tasks. A current number of stage tasks allocated to stage tracks from the storage unit to the cache is adjusted based on the computed number of stage tasks. A current number of destage tasks allocated to destage tracks from the cache to the storage unit is adjusted based on the computed number of destage tasks.

## Claims

### Claim at position 1

1-7. (canceled)

### Claim 8

A computer program product for allocating tasks to stage tracks from a storage unit to a cache and destage tracks from the cache to the storage unit, comprising a computer readable storage medium having computer readable program code embodied therein that when executed performs operations, the operations comprising:
providing a machine learning module that receives as input storage performance information related to processing of Input/Output (I/O) requests with respect to the storage unit;
determining an adjusted number of stage tasks;
determining an adjusted number of destage tasks;
retraining the machine learning module with the storage performance information to produce the adjusted number of stage tasks and the adjusted number of destage tasks; and
using the retrained machine learning module to produce a computed number of stage tasks to allocate to staging operations and a computed number of destage tasks to allocate to destaging operations.

### Claim 9 (depends on 8)

The computer program product of claim 8, wherein the determining the adjusted number of stage tasks and the adjusted number of destage tasks and retraining the machine learning module are performed in response to completing one of a staging operation and a destaging operation.

### Claim 10 (depends on 8)

The computer program product of claim 8, wherein the determining the adjusted number of stage tasks and the adjusted number of destage tasks comprises:
determining a margin of error of a threshold storage parameter value of a storage parameter and a current value of the storage parameter;
determining the adjusted number of stage tasks as a function of the computed number of stage tasks and the margin of error; and
determining the adjusted number of destage tasks as a function of the computed number of destage tasks and the margin of error.

### Claim 11 (depends on 8)

The computer program product of claim 8, wherein the operations further comprise:
determining a first margin of error of a first threshold storage parameter value of a first storage parameter and a first current value of the first storage parameter;
determining a second margin of error of a second threshold storage parameter value of a second storage parameter and a second current value of the second storage parameter, wherein the first and the second storage parameters comprise different performance metrics;
determining the adjusted number of stage tasks as a function of the computed number of stage tasks and at least one of the first margin of error and the second margin of error; and
determining the adjusted number of destage tasks as a function of the computed number of destage tasks and at least one of the first margin of error and the second margin of error.

### Claim 12 (depends on 11)

The computer program product of claim 11, wherein the function performs one of:
alternating using the first margin of error and the second margin of error to determine the adjusted number of stage tasks and the adjusted number of destage tasks during different iterations of performing the retraining the machine learning module; and
applying both the first margin of error and the second margin of error to the computed number of stage tasks and the computed number of destage tasks to determine the adjusted number of stage tasks and the adjusted number of destage tasks, respectively.

### Claim 13 (depends on 11)

The computer program product of claim 11, wherein a device adaptor transfers data between the storage unit and the cache, wherein the first storage parameter comprises device adaptor bandwidth, wherein the first threshold storage parameter value comprises an optimum adaptor bandwidth for the device adaptor and wherein the first current value of the first storage parameter comprises a current adaptor bandwidth of the device adaptor, wherein the second storage parameter comprises a response time of I/O requests to tracks in the storage unit, wherein the second threshold storage parameter value comprises a maximum acceptable response time for I/O requests and wherein the second current value of the second storage parameter comprises a current response time.

### Claim 14 (depends on 13)

The computer program product of claim 13, wherein the function uses the first margin of error and the second margin of error to increase the computed number of stage tasks and the computed number of destage tasks such that the adjusted number of stage tasks and the adjusted number of destage tasks are greater than the computed number of stage tasks and the computed number of destage tasks, respectively, in response to the first margin of error and the second margin of error being positive, wherein the function uses the first margin of error and the second margin of error to decrease the computed number of stage tasks and the computed number of destage tasks such that the adjusted number of stage tasks and the adjusted number of destage tasks are less than the computed number of stage tasks and the computed number of destage tasks, respectively, in response to the first margin of error and the second margin of error being negative.

### Claim 15 (depends on 8)

The computer program product of claim 8, wherein the storage unit is configured as Redundant Array of Independent Disk (RAID) ranks, wherein each of the RAID ranks is comprised of storage devices, wherein there is storage performance information for each of the RAID ranks, wherein the adjusted number of stage tasks and the adjusted number of destage tasks are determined separately for each of the RAID ranks, and wherein the machine learning module comprises one of at least one machine learning module that is retrained, for each RAID rank of the RAID ranks, with the storage performance information for the RAID rank to produce the adjusted number of stage tasks and the adjusted number of destage tasks for the RAID rank.

### Claim at position 10

16-17. (canceled)

### Claim 18

A system for allocating tasks to destaging and staging operations with respect to a storage unit, comprising:
a processor;
a cache implemented in at least one memory device;
a machine learning module; and
a computer readable storage medium having computer readable program code embodied therein that when executed by the processor performs operations, the operations comprising:
providing the machine learning module that receives as input storage performance information related to processing of Input/Output (I/O) requests with respect to the storage unit;
determining an adjusted number of stage tasks;
determining an adjusted number of destage tasks;
retraining the machine learning module with the storage performance information to produce the adjusted number of stage tasks and the adjusted number of destage tasks; and
using the retrained machine learning module to produce a computed number of stage tasks to allocate to staging operations and a computed number of destage tasks to allocate to destaging operations.

### Claim 19 (depends on 18)

The system of claim 18, wherein the operations further comprise:
determining a first margin of error of a first threshold storage parameter value of a first storage parameter and a first current value of the first storage parameter;
determining a second margin of error of a second threshold storage parameter value of a second storage parameter and a second current value of the second storage parameter, wherein the first and the second storage parameters comprise different performance metrics;
determining the adjusted number of stage tasks as a function of the computed number of stage tasks and at least one of the first margin of error and the second margin of error; and
determining the adjusted number of destage tasks as a function of the computed number of destage tasks and at least one of the first margin of error and the second margin of error.

### Claim 20 (depends on 19)

The system of claim 19, further comprising:
a device adaptor to transfer data between the storage unit and the cache, wherein the first storage parameter comprises device adaptor bandwidth, wherein the first threshold storage parameter value comprises an optimum adaptor bandwidth for the device adaptor and wherein the first current value of the first storage parameter comprises a current adaptor bandwidth of the device adaptor, wherein the second storage parameter comprises a response time of I/O requests to tracks in the storage unit, wherein the second threshold storage parameter value comprises a maximum acceptable response time for I/O requests and wherein the second current value of the second storage parameter comprises a current response time.

### Claim at position 14

21-22. (canceled)

### Claim 23

A method for allocating tasks to stage tracks from a storage unit to a cache and destage tracks from the cache to the storage unit, comprising:
providing a machine learning module that receives as input storage performance information related to processing of Input/Output (I/O) requests with respect to the storage unit;
determining an adjusted number of stage tasks;
determining an adjusted number of destage tasks;
retraining the machine learning module with the storage performance information to produce the adjusted number of stage tasks and the adjusted number of destage tasks; and
using the retrained machine learning module to produce a computed number of stage tasks to allocate to staging operations and a computed number of destage tasks to allocate to destaging operations.

### Claim 24 (depends on 23)

The method of claim 23, further comprising:
determining a first margin of error of a first threshold storage parameter value of a first storage parameter and a first current value of the first storage parameter;
determining a second margin of error of a second threshold storage parameter value of a second storage parameter and a second current value of the second storage parameter, wherein the first and the second storage parameters comprise different performance metrics;
determining the adjusted number of stage tasks as a function of the computed number of stage tasks and at least one of the first margin of error and the second margin of error; and
determining the adjusted number of destage tasks as a function of the computed number of destage tasks and at least one of the first margin of error and the second margin of error.

### Claim 25 (depends on 24)

The method of claim 24, wherein a device adaptor transfers data between the storage unit and the cache, wherein the first storage parameter comprises device adaptor bandwidth, wherein the first threshold storage parameter value comprises an optimum adaptor bandwidth for the device adaptor and wherein the first current value of the first storage parameter comprises a current adaptor bandwidth of the device adaptor, wherein the second storage parameter comprises a response time of I/O requests to tracks in the storage unit, wherein the second threshold storage parameter value comprises a maximum acceptable response time for I/O requests and wherein the second current value of the second storage parameter comprises a current response time.

### Claim 26 (depends on 8)

The computer program product of claim 8, wherein the storage unit comprises a Redundant Array of Independent Disk (RAID) rank of a plurality of RAID ranks, wherein each of the RAID ranks is comprised of storage devices, and wherein there is storage performance information for each of the RAID ranks, wherein a non-volatile storage (NVS) unit stores modified data in the cache, wherein the storage performance information for each RAID rank of the RAID ranks comprises at least one of:
number of tasks queued for staging operations;
speed of at least one storage device in which the RAID rank is stored;
overall NVS usage of tracks from all the RAID ranks;
NVS usage by the RAID rank;
maximum NVS usage allowed for the RAID rank;
RAID rank response time for processing I/O requests with respect to the storage unit;
device adaptor bandwidth utilized in a device adaptor that transfers data between the cache and the storage unit;
maximum device adaptor bandwidth that is available for transfer of data; and
optimum device adaptor bandwidth.

### Claim 27 (depends on 18)

The system of claim 18, wherein the determining the adjusted number of stage tasks and the adjusted number of destage tasks and retraining the machine learning module are performed in response to completing one of a staging operation and a destaging operation.

### Claim 28 (depends on 18)

The system of claim 18, wherein the determining the adjusted number of stage tasks and the adjusted number of destage tasks comprises:
determining a margin of error of a threshold storage parameter value of a storage parameter and a current value of the storage parameter;
determining the adjusted number of stage tasks as a function of the computed number of stage tasks and the margin of error; and
determining the adjusted number of destage tasks as a function of the computed number of destage tasks and the margin of error.

### Claim 29 (depends on 19)

The system of claim 19, wherein the function performs one of:
alternating using the first margin of error and the second margin of error to determine the adjusted number of stage tasks and the adjusted number of destage tasks during different iterations of performing the retraining the machine learning module; and
applying both the first margin of error and the second margin of error to the computed number of stage tasks and the computed number of destage tasks to determine the adjusted number of stage tasks and the adjusted number of destage tasks, respectively.

### Claim 30 (depends on 20)

The system of claim 20, wherein the function uses the first margin of error and the second margin of error to increase the computed number of stage tasks and the computed number of destage tasks such that the adjusted number of stage tasks and the adjusted number of destage tasks are greater than the computed number of stage tasks and the computed number of destage tasks, respectively, in response to the first margin of error and the second margin of error being positive, wherein the function uses the first margin of error and the second margin of error to decrease the computed number of stage tasks and the computed number of destage tasks such that the adjusted number of stage tasks and the adjusted number of destage tasks are less than the computed number of stage tasks and the computed number of destage tasks, respectively, in response to the first margin of error and the second margin of error being negative.

### Claim 31 (depends on 18)

The system of claim 18, wherein the storage unit is configured as Redundant Array of Independent Disk (RAID) ranks, wherein each of the RAID ranks is comprised of storage devices, wherein there is storage performance information for each of the RAID ranks, wherein the adjusted number of stage tasks and the adjusted number of destage tasks are determined separately for each of the RAID ranks, and wherein the machine learning module comprises one of at least one machine learning module that is retrained, for each RAID rank of the RAID ranks, with the storage performance information for the RAID rank to produce the adjusted number of stage tasks and the adjusted number of destage tasks for the RAID rank.

### Claim 32 (depends on 23)

The method of claim 23, wherein the storage unit comprises a Redundant Array of Independent Disk (RAID) rank of a plurality of RAID ranks, wherein each of the RAID ranks is comprised of storage devices, and wherein there is storage performance information for each of the RAID ranks, wherein a non-volatile storage (NVS) unit stores modified data in the cache, wherein the storage performance information for each RAID rank of the RAID ranks comprises at least one of:
number of tasks queued for staging operations;
speed of at least one storage device in which the RAID rank is stored;
overall NVS usage of tracks from all the RAID ranks;
NVS usage by the RAID rank;
maximum NVS usage allowed for the RAID rank;
RAID rank response time for processing I/O requests with respect to the storage unit;
device adaptor bandwidth utilized in a device adaptor that transfers data between the cache and the storage unit;
maximum device adaptor bandwidth that is available for transfer of data; and
optimum device adaptor bandwidth.

### Claim 33 (depends on 23)

The method of claim 23, wherein the determining the adjusted number of stage tasks and the adjusted number of destage tasks and retraining the machine learning module are performed in response to completing one of a staging operation and a destaging operation.

### Claim 34 (depends on 23)

The method of claim 23, wherein the determining the adjusted number of stage tasks and the adjusted number of destage tasks comprises:
determining a margin of error of a threshold storage parameter value of a storage parameter and a current value of the storage parameter;
determining the adjusted number of stage tasks as a function of the computed number of stage tasks and the margin of error; and
determining the adjusted number of destage tasks as a function of the computed number of destage tasks and the margin of error.

### Claim 35 (depends on 24)

The method of claim 24, wherein the function performs one of:
alternating using the first margin of error and the second margin of error to determine the adjusted number of stage tasks and the adjusted number of destage tasks during different iterations of performing the retraining the machine learning module; and
applying both the first margin of error and the second margin of error to the computed number of stage tasks and the computed number of destage tasks to determine the adjusted number of stage tasks and the adjusted number of destage tasks, respectively.

### Claim 36 (depends on 35)

The method of claim 35, wherein the function uses the first margin of error and the second margin of error to increase the computed number of stage tasks and the computed number of destage tasks such that the adjusted number of stage tasks and the adjusted number of destage tasks are greater than the computed number of stage tasks and the computed number of destage tasks, respectively, in response to the first margin of error and the second margin of error being positive, wherein the function uses the first margin of error and the second margin of error to decrease the computed number of stage tasks and the computed number of destage tasks such that the adjusted number of stage tasks and the adjusted number of destage tasks are less than the computed number of stage tasks and the computed number of destage tasks, respectively, in response to the first margin of error and the second margin of error being negative.

### Claim 37 (depends on 23)

The method of claim 23, wherein the storage unit is configured as Redundant Array of Independent Disk (RAID) ranks, wherein each of the RAID ranks is comprised of storage devices, wherein there is storage performance information for each of the RAID ranks, wherein the adjusted number of stage tasks and the adjusted number of destage tasks are determined separately for each of the RAID ranks, and wherein the machine learning module comprises one of at least one machine learning module that is retrained, for each RAID rank of the RAID ranks, with the storage performance information for the RAID rank to produce the adjusted number of stage tasks and the adjusted number of destage tasks for the RAID rank.

### Claim 38 (depends on 23)

The method of claim 23, wherein the storage unit comprises a Redundant Array of Independent Disk (RAID) rank of a plurality of RAID ranks, wherein each of the RAID ranks is comprised of storage devices, and wherein there is storage performance information for each of the RAID ranks, wherein a non-volatile storage (NVS) unit stores modified data in the cache, wherein the storage performance information for each RAID rank of the RAID ranks comprises at least one of:
number of tasks queued for staging operations;
speed of at least one storage device in which the RAID rank is stored;
overall NVS usage of tracks from all the RAID ranks;
NVS usage by the RAID rank;
maximum NVS usage allowed for the RAID rank;
RAID rank response time for processing I/O requests with respect to the storage unit;
device adaptor bandwidth utilized in a device adaptor that transfers data between the cache and the storage unit;
maximum device adaptor bandwidth that is available for transfer of data; and
optimum device adaptor bandwidth.
